knowledge

Blueprint

How knowledge is held, governed, and exchanged

A brain is a locally owned, policy governed, append only audited folder of plain text. Blueprint specifies the structure and governance of one brain; its six companions specify what the brain is for, who may decide, how knowledge moves between brains, how a model may process it, what tooling leaves behind, and what an agent that learns is allowed to keep.

Draft. Nothing in this repository is adopted until it lands as a decision record. Conformance claims against a draft are premature.
v1.0draftApache 2.0

stack

The fourth member of the stack

The agent stack governs agents as digital workers; Blueprint governs the brains they work with. SAGA moves an agent. Blueprint moves a thought.

ABR what agents deserve · agent-rights.org
DERP what the runtime must provide · derp-spec.dev
SAGA how an agent is represented · saga-standard.dev
Blueprint how knowledge is held, governed, and exchanged

documents

Seven documents, one brain

One brain, three boundaries, one authority model, one reason. Knowledge crosses to another brain under SPEAK, to a model under CONFIDE, and into the tooling under TRACE. Every crossing is an act, DEFER says who may authorize it, and POLARIS says whether the brain will do it at all.

01

BLUEPRINT

structure

A single brain, in ten layers.

The folder itself: identity, records, policy, ledger, and the layers that make a directory of plain text into an owned, audited body of knowledge.

02

POLARIS

purpose

Purpose and alignment.

What the brain exists for, and the refusals, obligations, loyalties, and standards that narrow every act. Highest precedence to forbid, none to permit.

03

DEFER

authority

Delegated authority.

Consequence classes, authority envelopes, signed grants, routing, escalation, and decision records. Who may decide, bounded by what.

04

SPEAK

exchange

Brain-to-brain exchange.

How knowledge moves between brains without merging them: signed, classified, reconciled, auditable.

05

CONFIDE

inference

Inference governance.

Which providers may see which records, under what evidence, logged to a hash-chained ledger. Processing without leaving the owner's control.

06

TRACE

residue

Tooling residue.

Harness registration, artifact classes, session anchors, sealing, and retention: the crossing that happens on every operation, and the one nobody declared.

07

RETAIN

agent brains

Whose brain an agent has.

When accumulated agent state is residue, when it is a domain you own, and when it is a brain someone else owns; plus what an agent keeps when the engagement ends.

principles

Two rules that carry the design

Precedence is asymmetric, and the asymmetry is the point. POLARIS holds the highest precedence to forbid an act and no precedence to permit one. Purpose may narrow what a brain will do. It may never widen it, satisfy another document's check, or excuse a failure.

There are no sub-brains. Containment and boundary are mutually exclusive: if you can read it at will, the agent has a folder; if you cannot, it is not inside you. Whoever holds the signing key owns the refusal, and a brain you can sign for is a folder with aspirations. What looks like a hierarchy of brains is a set of peers under asymmetric agreements plus an authority graph that DEFER already governs.